nc://pareto

Pareto Inference: GLM 5.3 Flash, DeepSeek V4.1 Flash and MiMo V2.6 Flash, pay per token. A Nodecode rewrite of pareto-inference/hermes-pareto-provider by pareto-inference.

Pareto speaks the OpenAI chat wire at https://api.paretoinference.com/v1. With this cell, /connect offers Pareto Inference, /models lists its three models, and a turn goes out in the form Pareto takes:

• the model is named vendor/model on the wire, so the picker's bare glm-5.3-flash is sent as z-ai/glm-5.3-flash
• reasoning_effort is sent top-level in Pareto's spelling: the picker's rungs go through as they are, ultra becomes max, and turning reasoning off sends none (Nodecode's own off rung is the wire's none)
• every request streams, which is what Pareto requires: a plain call is answered HTTP 400 streaming_only, and Nodecode's chat lane only streams

Pareto bills per token; its pricing is at docs.paretoinference.com/pricing.

Get it

Tell the agent "add the pareto cell", or run nodecode add nc://pareto.

Set it up

Make an API key at paretoinference.com/dashboard, then pick Pareto Inference in /models and paste the key when the picker asks, or set PARETO_API_KEY in the environment. The address is optional, for a relay or a proxy:

{
  "pareto": {
    "base_url": "https://api.paretoinference.com/v1"
  }
}

Use it

Ask for one of the models — /models lists what Pareto Inference serves — then talk as usual. /think offers the rungs Pareto takes (off through max); one it does not take is left out of the request, so Pareto's own per-model default applies.

Test

From this repository's root:

.github/test-cell.sh providers/nodecode-pareto

MIT licensed. See NOTICE.